[DEBATE] : "If South Africa were a single village (sic) with only 100 inhabitants, what would it look like?"

[Patrick Bond]

A caution on this, comrades:

Sean Jacobs wrote:
> ... If the village only consisted of 100 adults….i.e. those who are 16 
> years and older…
> • 42 are employed (full-time or part-time)
> • 26 are unemployed and looking for work
> • 49 are poor (the total household income per month is below R2 499)
> • 4 earn incomes of R300 000 per year or higher
> • 14 earn incomes of R100 000 per year or higher
> • 25 earn incomes of R50 000 per year or higher
> ...
> Sources:
> Ipsos Markinor Khayabus, Demographic Detail. November 2007.
> Ipsos Markinor. Socio-Political Trends. November 2007.
> BMR. Income and Expenditure Model. 2008 Update.
> AMPS 2007.

There are some serious questions about the poverty/inequality data 
emanating from both StatsSA and AMPS. The most recent paper I've found 
on this problem is by the brilliant marxist economist Charles Meth 
(intro below - full available on request offlist to pbond at mail.ngo.za). 
There's a pro-government bias both in data and the methodological 
approach used by the main pro-government researcher, Servaas van der 
Berg of Stellenbosch. Meth does another demolition job here (as well as 
one in the CCS special issue of Africanus, available at
http://www.nu.ac.za/ccs/files/africanus_1.pdf)...


Flogging a dead horse:
Attempts by van der Berg et al to measure changes in poverty and inequality

by
Charles Meth
Southern Africa Labour and Development Research Unit

Abstract
This paper seeks an explanation for the large differences in the extent 
and severity of poverty published respectively in van der Berg et al 
(2005: 2007a) and Meth (2006b). Headcounts in 2004 suggested by van der 
Berg et al (2007a) exceed by five million, those reported by
(Meth, 2006b). Household survey respondents often under-report income 
(and expenditure). To address this, it is common (if not necessarily 
wise) to scale household survey income means until the grossed-up survey 
income totals are approximately the same as those yielded by the 
national accounts. The apparent reason for the differences between our 
respective poverty estimates lies in the poor quality of the income 
estimates in the surveys used by van der Berg et al as primary data 
source for estimating income distributions (by race). Scaling these 
survey estimates to make them consistent with the national accounts, it 
is argued, causes them to under-estimate the extent and severity of the 
poverty problem. As part of their analysis of changes in the welfare of 
Africans in South Africa since the advent of democracy (and in support 
of their claim that poverty has fallen), van der Berg et al attempt to 
measure changes in the racial shares of remuneration. The present paper 
ends with a brief examination of some of the problems of doing so using 
Statistics South Africa household surveys (the Labour Force Surveys) as 
primary data source. Welcomed by government because of the apparent 
progress they report in the fight against poverty, the possible 
consequences for anti-poverty policy (and for the poor) of the van der 
Berg et al figures being wrong are non-trivial.

Introduction

In 2005, Professor van der Berg and his colleagues in the University of 
Stellenbosch published a set of poverty estimates which have proved to 
be enormously influential, not least because they posit a substantial 
reduction in the severity of poverty in the period 2000-2004
(van der Berg et al, 2005). Reworking the estimates has led to the 
publication of a set estimates that register even lower headcounts
(van der Berg et al, 2007a). In response to the claims made in the 2005 
paper, I have written two papers (Meth, 2006a and 2006b), the first of 
which uses the expenditure estimates in the Labour Force Surveys
(LFSs). The second makes use of the income figures in the LFSs. Both 
efforts discover higher poverty headcounts and lower rates of poverty 
reduction than those reported by van der Berg and his co-authors. In my 
2006b paper, using the same poverty line as van der Berg et al
(2005), I estimated that there were about 18 million people below the 
poverty line in 2004. Of them, I argued: “… 14 million lived in 
workerless households (most containing working age people, but in which 
nobody had employment). These zero-income (from employment, that is) 
households survived on a mix of social grants and/or remittances. Among 
them were about 1.8 million people in households receiving no incomes at 
all in the survey reference period, subsisting, we know not how. The 
remaining four million people below the poverty line were located in 
households containing about 800 000 workers. Although the bulk of 
poverty is caused by unemployment, the problem of the working poor still 
looms fairly large.” (Abstract) With a poverty line of R250 per capita 
per month in 2000 prices, the original paper by van der Berg et al that 
made use of the AMPS (All Media and Products Study) data, had headcounts 
of 16.2, 18.5 and 15.4 million in 1993, 2000 and 2004 respectively 
(2005, Table 2, p.17). In the most recent offering, headcounts in the 
same three years fall to 13.4, 16.3 and 13.1 million (2007a, Table 2, 
p.19). The increase in the headcount between 1993 and 2000 is slightly 
higher, but expansion of the social grant system (and whatever job and 
real income growth there was) has roughly the same absolute impact as 
before, knocking about 3.1 million off the headcount between 2000 and
2004. The poverty line is the same (p.19). As noted below, apart from a 
short reference to ‘small improvements’ in the technique for estimating 
the distribution of wage income, there is no explanation for the 
substantial differences between their 2005 and 2007a headcount 
estimates. So, not only do they repeat the claim that poverty dropped by 
three million between 2000 and 2004; their latest estimates of the 
headcounts for 2000 and 2004 are now some two million lower than their
2005 estimates. In academic terms, of course, the fact that my estimates 
are higher than theirs is neither here nor there – my figures could 
equally well be wrong. A problem arises, though, if they are not. The 
van der Berg et al poverty findings have attracted a huge amount of 
attention (and publicity) – government has made frequent use of them to 
show that anti-poverty policies are succeeding, they almost certainly 
form an important part of the basis for the government assertion (made 
on numerous occasions) that the goal of halving poverty by 2014 will be 
met. Treasury officials have tried to dismiss the (previous) differences 
between our findings as trivial – these new lower headcounts make that 
stance even less defensible than before – the difference between our 
estimates of the poverty headcount in 2004 is now almost five million! 
By their own admission, their latest estimates of the poverty levels are 
“artificially low” (van der Berg et al, 2007a, Abstract). This admission 
marks a shift from their earlier stance, where they claimed that: “The 
assumptions used throughout the study are those likely to yield the 
lowest estimates of poverty reduction that the national accounts data 
support. Thus our estimates are also purposely biased towards recording 
the least rather than the most likely estimates of income growth for the 
black population, since this group contains the majority of the poor. 
Also, despite reservations that we have about some spikes in the data 
obtained from official surveys (in particular the high levels of wages 
recorded for particularly the black population in 1995 and the low 
levels recorded for 2000), we do not adjust for these and instead use 
the most conservative estimates of black wages. Thus our estimates 
probably overstate poverty compared to estimates that also adjust data 
to be commensurate with the national accounts.” (van der Berg et al, 
2005, p.4, emphasis in original) Recognising, as they could hardly fail 
to do, the essentially arbitrary character of poverty lines, the van der 
Berg oeuvre is replete with references to the need to uncover trends, 
presumably in preference to a concentration on absolute levels per se. 
They cite, for example, an argument in defence of the adjustment of 
survey means using national accounts data, which speaks of the need to 
select methods of treating data which: “… minimizes errors, especially 
errors in trends, because that is an important variable of interest.”
(van der Berg et al, 2007a, p.9) As I have pointed out elsewhere (Meth, 
2006a, p.2), and as they themselves recognise, talk of trends is 
somewhat misleading. In their own words, “… social assistance is nearing 
the boundaries of its ability to alleviate poverty.” (van der Berg et 
al, 2005, p.3). The South African government is firmly set against 
extension of the social grant system (the major cause of such poverty 
reduction as has taken place since 2000) beyond its present limits 
(Meth, 2007b, pp.17ff). Unless rapid job-creating growth among the poor 
takes place, the trend they uncover will soon be no more. In previous 
encounters with the van der Berg et al results, although I have hinted, 
in personal communications, in a seminar setting,2 and in my own 
writings on the topic (see Meth, 2006a, pp.55-56), at a potentially 
fatal flaw at the heart of their workmanship, I have steered clear of 
any detailed engagement with their method. The release of their latest 
figures, means that it is no longer advisable simply to treat the causes 
of the differences between our results as if they were no more 
consequential than a debate about how many angels could dance on the 
head of a pin. Accordingly, therefore, the present paper attempts to get 
to the heart of the differences between our results. The paper commences 
with a quibble about the way in which van der Berg et al (2007a) attempt 
to smooth over these differences. The central section of the paper is 
devoted to an exposition of that part of their methodology within which 
the problem is suspected to lie. The investigation closes in on the 
relevant bits of the AMPS survey questionnaire for the year 2004, 
analysis of which suggests that it is the form the income question takes 
that explains the differences between us. In passing, comment is offered 
on the difficulties of estimating racial mean incomes at a national 
level. Since van der Berg et al make great play of rising African shares 
of remuneration, some attention is paid in the final section of the 
paper to the difficulties of creating reliable estimates of the relative 
magnitudes of the shares of the different race groups.